Suche Bilder Maps Play YouTube News Gmail Drive Mehr »
Anmelden
Nutzer von Screenreadern: Klicke auf diesen Link, um die Bedienungshilfen zu aktivieren. Dieser Modus bietet die gleichen Grundfunktionen, funktioniert aber besser mit deinem Reader.

Patentsuche

  1. Erweiterte Patentsuche
VeröffentlichungsnummerUS20080270109 A1
PublikationstypAnmeldung
AnmeldenummerUS 12/132,401
Veröffentlichungsdatum30. Okt. 2008
Eingetragen3. Juni 2008
Prioritätsdatum16. Apr. 2004
Auch veröffentlicht unterUS8666725, US8977536, US20060015320, WO2005099422A2, WO2005099422A3
Veröffentlichungsnummer12132401, 132401, US 2008/0270109 A1, US 2008/270109 A1, US 20080270109 A1, US 20080270109A1, US 2008270109 A1, US 2008270109A1, US-A1-20080270109, US-A1-2008270109, US2008/0270109A1, US2008/270109A1, US20080270109 A1, US20080270109A1, US2008270109 A1, US2008270109A1
ErfinderFranz Josef Och
Ursprünglich BevollmächtigterUniversity Of Southern California
Zitat exportierenBiBTeX, EndNote, RefMan
Externe Links: USPTO, USPTO-Zuordnung, Espacenet
Method and System for Translating Information with a Higher Probability of a Correct Translation
US 20080270109 A1
Zusammenfassung
A system with a nonstatistical translation component integrated with a statistical translation component engine. The same corpus may be used for training the statistical engine and also for determining when to use the statistical engine and when to use the translation component. This training may use probabilistic techniques. Both the statistical engine and the translation components may be capable of translating the same information, however the system determines which component to use based on the training. Retraining can be carried out to add additional components, or when after additional translator training.
Bilder(4)
Previous page
Next page
Ansprüche(24)
1. A method comprising:
translating information from a first language to a second language using at least two translation components, each of the at least two translation components capable of translating equivalent phrases; and
automatically selecting a preferred component from the at least two translation components, the preferred component providing a translation having a highest probability of being correct.
2. A method as in claim 1, further comprising defining a feature function that indicates when to use the at least two translation components.
3. A method as in claim 1, wherein automatically selecting the preferred component comprises:
obtaining a phrase to be translated; and
computing a probability of being correct in translating the phrase for each of the at least two translation components.
4. A method as in claim 3, further comprising:
detecting variants of a translation of the phrase; and
accepting the translation of the phrase as being proper if the translation of the phrase is equivalent to one of the variants.
5. A method as in claim 4, further comprising annotating a training corpus based on the variants to form an annotated training corpus.
6. A method as in claim 1, wherein a first component of the at least two translation components comprises a statistical translator and a second component of the at least two translation components comprises a nonstatistical translator for proper nouns.
7. A method as in claim 1, wherein a first component of the at least two translation components comprises a statistical translator and a second component of the at least two translation components comprises a nonstatistical translator for names.
8. A method as in claim 1, wherein a first component of the at least two translation components comprises a statistical translator and a second component of the at least two translation components comprises a nonstatistical translator for numbers.
9. A method as in claim 1, further comprising automatically re-selecting a second preferred component from the at least two translation components in response to an occurrence.
10. A method as in claim 9, wherein the occurrence comprises adding an additional nonstatistical translation component to the at least two translation components.
11. A method as in claim 9, wherein the occurrence comprises improving at least one of the at least two translation components.
12. A method as in claim 1, further comprising:
training a format of an output of a machine translation system based on a training corpus; and
allowing at least one of a plurality of different formats to be selected from within the training corpus.
13. A system comprising:
at least two translating parts, each of the at least two translating parts operational to translate information from a first language to a second language and each of the at least two translating parts capable of translating equivalent phrases; and
a classifier part operational to automatically select a preferred component from the at least two translation components, the preferred component providing a translation having a highest probability of being correct.
14. A system as in claim 13, further comprising a training corpus.
15. A system as in claim 14, further comprising an output module that formats an output based on the training corpus.
16. A system as in claim 13, further comprising a variant detector that detects variants of a translated phrase and accepts the translated phrase as being proper if the translation phrase is one of the variants.
17. A system as in claim 16, further comprising an annotated training corpus based, at least in part, on the variants.
18. A system as in claim 13, wherein a first part of the at least two translating parts comprises a statistical translator component and a second part of the at least two translating parts comprises a nonstatistical translator component for proper nouns.
19. A system as in claim 13, wherein a first part of the at least two translating parts comprises a statistical translator component and a second part of the at least two translating parts comprises a nonstatistical translator component for names.
20. A system as in claim 13, wherein a first part of the at least two translating parts comprises a statistical translator component and a second part of the at least two translating parts comprises a nonstatistical translator component for numbers.
21. A system as in claim 13, further comprising a feature part that indicates when to use the at least two translation components.
22. A system as in claim 13, wherein the classifier part comprises a probabilistic classifier.
23. A system as in claim 13, further comprising at least one additional translating part.
24. A computer readable storage medium having embodied thereon a program, the program executable by a processor to perform a method, the method comprising:
translating information from a first language to a second language using at least two translation components, each of the at least two translation components capable of translating equivalent phrases; and
automatically selecting a preferred component from the at least two translation components, the preferred component providing a translation having a highest probability of being correct.
Beschreibung
    CROSS-REFERENCE TO RELATED APPLICATION
  • [0001]
    This application is a divisional and claims the priority of U.S. patent application Ser. No. 11/107,304, entitled “Selection and Use of Nonstatistical Translation Components in a Statistical Machine Translation Framework,” filed on Apr. 15, 2005, which claims the benefit of U.S. Provisional Patent Application No. 60/562,774, filed on Apr. 16, 2004, the subject matter of which are hereby incorporated by reference.
  • FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT
  • [0002]
    The U.S. Government may have certain rights in this invention pursuant to Grant No. N66001-00-1-8914 awarded by DARPA.
  • BACKGROUND
  • [0003]
    Statistical machine translation automatically learns how to translate using a training corpus. The learned information can then be used to translate another, “unknown” text, using information that the machine learned from the training operation.
  • [0004]
    However, current statistical machine translation models are typically not suited for certain types of expressions, e.g., those where statistical substitution is not possible or feasible. For example, the current state of statistical machine translation systems does not allow translating Chinese numbers into English until the numbers have been seen and the correct translation has been learned. Similar issues may exist for translations of names, dates, and other proper nouns.
  • [0005]
    In addition, it may be desirable to conform a machine translation output to certain formats. The most desirable format may be different than the training corpus, or inconsistent within the training corpus. As an example, Chinese names may be present in a training corpus with the family name first, followed by the surname. However, it is more conventional to print the translation in English with the first name first. This may make it desirable to change the output in order to deviate what was seen in the parallel training data.
  • [0006]
    Certain modern statistical machine translation systems have integrated a rule based translation component for things like numbers and dates. There have also been attempts to combine statistical translation with other full sentence machine translation systems by performing an independent translation with the different systems and deciding which of the systems provides a better translation.
  • SUMMARY
  • [0007]
    An aspect of the present system is to integrate non-statistical translation components, along with statistical components, to use certain components for certain kinds of translation. An aspect allows training to determine when it is desirable to use different components for different parts of the translation operation.
  • [0008]
    The techniques described herein use a parallel training corpus. The system may automatically learn from the corpuses where entity translation component or components are likely to produce or better translations. This system can automatically learn a confidence factor for different entity translation components in specific contexts. Therefore, this approach can also adapt to unreliable entity translation components.
  • BRIEF DESCRIPTION OF THE DRAWINGS
  • [0009]
    These and other aspects will now be described in detail with reference to the accompanying drawings, wherein:
  • [0010]
    FIG. 1 shows a block diagram of the translation system;
  • [0011]
    FIG. 2 shows a flowchart of training a classifier that determines when to use different components for different translations; and
  • [0012]
    FIG. 3 shows a flowchart of operation using multiple translation components.
  • DETAILED DESCRIPTION
  • [0013]
    The present system describes integration of non-statistical machine translation components into a statistical machine translation framework. This is done by using a processing device to determine automatically which parts of an input string should use a “baseline” machine translation system, and which parts should use another entity translation component or components, referred to herein as the translation component.
  • [0014]
    FIG. 1 illustrates an exemplary hardware device which may execute the operation that is described with reference to the flowcharts of FIGS. 2 and 3. For the application of language translation, a processing module 100 receives data from various sources 105. The sources may be parallel corpora of multiple language information. Specifically, the sources may include translation memories, dictionaries, glossaries, Internet information, parallel corpora in multiple languages, non-parallel corpora in multiple languages having similar subject matter, and human-created translations. The processor 100 processes this information to produce translation parameters which are output as 110. The translation parameters are used by language engine 120 in making translations based on input language 130. In the disclosed embodiment, the language engine 120 includes a statistical engine 123, and at least one translation component 124. The language engine translates from a first language to a second language. However, alternatively, the speech engine can be any engine that operates on strings of words, such as a language recognition device, a speech recognition device, a machine paraphraser, natural language generator, modeler, or the like.
  • [0015]
    The processor 100 and speech engine 120 may be any general purpose computer, and can be effected by a microprocessor, a digital signal processor, or any other processing device that is capable of executing the operations described herein.
  • [0016]
    The flowcharts described herein can be instructions which are embodied on a machine-readable medium such as a disc or the like. Alternatively, the flowchart can be executed by dedicated hardware, or by any known or later discovered processing device.
  • [0017]
    The translation component 124 can be any existing translation component of any type, including a rule-based translator, or any other kind of machine translation component. Such translation components may be capable of translating many different kinds of information from one language to another.
  • [0018]
    In the embodiment, translation component 124 is used to translate only a portion of the information that it is capable of translating. For example, the translation component may be capable of translating standard two or three character Chinese names. This may apply to many different Chinese size strings. This may include, for example, certain strings which are not actually names. One aspect of the system is to identify the portions with are desired to be translated by the translation component. For example, in the above example, the component must determine how to identify the Chinese names in text, and then to translate those names using the component 124. Other Chinese language information is translated using the statistical engine 123.
  • [0019]
    Another aspect is detects whether the translation component uses a complete and/or accurate rule set. For example, if the rule set for the translation component 124 for a specific translation is incomplete, then the engine 120 will consider using instead the baseline statistical machine translation part 123.
  • [0020]
    Using the above example, therefore, the goal is to identify Chinese names where the translation component 124 produces a correct translation. The translation component can therefore be used for entities that are not actually person names and can be translated; for example, company names that are constructed like person names.
  • [0021]
    Therefore, the training of the machine trains not only the statistical machine translation, but also trains when to use the statistical machine translation. The translator is give a source sentence in a source language, for example Chinese, which is to be translated into a target language, for example English. Among all possible target sentences, the machine may choose the sentence with the highest probability
  • [0000]
    e ^ = arg max e { Pr ( e | f ) } ( 1 )
  • [0022]
    Where the symbol Pr(.) represents general probability distributions with no, or virtually no, assumptions, argmax, denotes the search for the output sentence in the target language, and e is the sentence.
  • [0023]
    The posterior probability is modeled using a log linear model. This framework produces a set of M feature functions hm(e,f), m−1 . . . M.
  • [0024]
    Each feature function M also has a model parameter λm, where m=1 . . . M.
  • [0025]
    The direct translation probability is given by:
  • [0000]
    Pr ( e | f ) = p λ j M ( e | f ) = exp [ m = 1 M λ m h m ( e , f ) ] e exp [ m = 1 M λ m h m ( e , f ) ] ( 2 ) ( 3 )
  • [0000]
    information may be translated by developing feature functions that capture the relevant properties of the translation task. These basic feature functions may include the alignment template approach described in “Discriminative Training And Maximum Entropy Models For Statistical Machine Translation”, Och and Ney 2002, proceedings of the 40th annual meeting of the Association for computational linguistics. This translation model segments the input sentence into phrases, translates these phrases, and reorders the translations into the target language.
  • [0026]
    Another possible feature function is a trigram language model. The feature functions may be trained using the unsmoothed maximum BLEU criterion, described in minimum error rate training in statistical machine translation (Och, 2003).
  • [0027]
    Training procedures for obtaining alignment templates is described in (Och 1999). Computation of word alignment in the parallel training corpus may use an expectation maximization technique, and a statistical alignment technique. See for example (Och and Ney 2003). This word alignment forms the basis for computing the probabilistic phrase to phrase translation lexicon p(e|f), which is used to store the translation of the phrase.
  • [0028]
    The translation component 124 is a machine translation system or module that can translate specific source language expressions into corresponding target language expressions. The translation component may provide the translation that is “best”, or may alternatively combine a candidate list of translation possibilities.
  • [0029]
    Different environments may use different translations. For example, the translation components may include:
      • a Chinese name translation—this translation component is a simple rule-based translation component that operates for two and three character Chinese names. This is done by applying the Pinyin rules to Chinese characters that frequently occur as parts of names, to identify and translate those Chinese names.
      • Number translation—this translation component performs a rule-based translation of Chinese numbers, percentages, and time expressions. It operates by determining such numbers percentages and time expressions, and translating them using rules.
      • Date translation—this translation component translates the expressions. One example is Nov. 2, 1971. The translation component will automatically translate this to the proper language.
  • [0033]
    An important issue is integration of these components with the statistical translator and training of when to use which one.
  • [0034]
    An ideal translation component provides no wrong translations at all. It provides the set of all correct translations for a given substring. Real world translation components make errors, and provide incorrect translations. For example, the Chinese name entity translation component frequently generates wrong translations when applied to Korean names. Certain expressions cannot be easily translated by the component. For example the date translator may provide 27 days, or the 27th as potential translations of the same characters. Only one of the two is correct for a specific context. Proper integration of the statistical translator with a translation component, therefore, requires learning/training when to use each of the components, and also training of the proper format to output.
  • [0035]
    FIG. 2 shows a flowchart showing how to learn automatically from a set of translation components in a parallel corpus, and to determine automatically which of the statistical engine 123, or the translation component 124, should be used to translate the source language string.
  • [0036]
    At 200, a translation component is annotated to list each substring that is capable of being translated by a translation component. Note that there may be one or many different translation components. The annotated corpus indicates which words/portions in the corpus can be translated with any of those translation components. That is done by determining words in the source language, that have a translation, via a translation component, actually occurring in the corresponding target language segment.
  • [0037]
    In an implementation, this may be carried out by applying all the translation components to all the source language substrings of the training corpus. The target language corpus may be used to determine if the training components has produced a correct translation.
  • [0038]
    A variant filter at 210 is used to attempt to prevent different forms of the same word from being rejected. The translation component at 200 may classify a correct translation as being wrong if the parallel training corpus is used as a variant of what the training component has proposed. The variant filter may analyze all or many of the possible translations. For example, all of the following strings: a thousand, one thousand or 1000, refer to the same number. Any of these is the correct translation of the Chinese word for “thousand”. The variant filter may allow any of these translations to be accepted.
  • [0039]
    It may be desirable to provide enough precision in the translation component to avoid negative instances as being misclassified as positive instances.
  • [0040]
    At 220, the annotated corpus is used for classifier training. A probabilistic classifier is trained based on the data. The classifier may be part of the processor 100. The classifier determines, for each source language sub string, and its source language context, if the translation component has actually produced a correct translation, or not a correct translation.
  • [0041]
    In operation, given a large parallel training corpus, a very large annotated corpus may be automatically generated. For language pairs like Chinese/English and Arabic/English, there may be readily available parallel corpora of more than 100 million words. Human-annotated training corpora are typically much smaller, e.g., they may be rarely less than larger than one million words.
  • [0042]
    Another aspect is that the automated annotation may be directly oriented toward the ultimate goal which is to use a certain translation component to produce correct translations. As a result, those instances for which the translation component produces a wrong translation may be annotated as negative instances.
  • [0043]
    When the translation component 124 is improved via increased coverage or improved quality of translation, an annotated corpus can be automatically regenerated at 230. The model may then be retrained to detect when to use the improved training corpus. Similarly, re-training can occur when the statistical database 123 is improved, when a new translation component is added, or when some other situation occurs.
  • [0044]
    This allows integration of different training components that each translate the same kind of instructions. The system learns automatically in this way when to trust which translation component. This allows automatic determination of which are acceptable and not acceptable translation components for particular words in particular contexts.
  • [0045]
    Mathematically speaking, to determine if the certain source language substrate of a source language string can be translated with the correct translation component to produce the translation, a model can be trained according to:
  • [0000]

    p(c|fj 1 j 2 ,fj 1 −2 j 1 −1,fj 2 +1 j 2 +2,TCn,e1 1)  (4)
  • [0046]
    Where fj represent substrings of a source language string; TCn is a specific translation component, and c stands for the two situations where “the translation component produces the correct translation” or “the translation component does not produce the correct translation”. A standard maximum entropy model described by Berger 96 may be used that uses each single dependent variable in equation 4 as a feature, is combined with the class c.
  • [0047]
    Different classifier models may be used for this framework, besides the maximum entropy classifier. A maximum entropy classifier may obtain probabilities which can be reasonably compared for different substrings.
  • [0048]
    FIG. 3 shows the overall operation of using the engine. The classifier is trained at 300, using the flowchart of FIG. 2. Once the classifier is trained in this way, the translation component is integrated into the overall process of the phrase based statistical machine translation system at 310. Each sub string of the text to be translated is analyzed at 320. The operation computes the probability that the translation component will produce a correct translation. A filter at 330 uses a threshold pmin to filter those cases where the probability of correct translation is too low. The resulting set of named entities is then used as an additional phrase translation candidates. These are hypothesized in search together with the phrases of the baseline statistical machine translation system at 340.
  • [0049]
    The statistical machine translation system balances between the use of translation component phrases and baseline system phrases. This may be defined by an additional feature function which counts the number of translation component phrases that are used. This may be stored as a variable referred to as TC-PENALTY. Other feature functions, such as a language model, or a reordering model, may also score those phrases.
  • [0050]
    Another aspect may enforce the use of translation component phrases if the corresponding source language sub string is rarely seen.
  • [0051]
    The translation component may also be integrated into the word alignment process between the parallel corpora. This may be done to improve word alignment accuracy during training. This procedure may automatically detect whether the translation component is trained sufficiently to be reliable. Once the translation components is sufficiently reliable, that information can be used to constrain the word alignment generated in the training procedure better alignment between the two languages may be obtained by using the translation components for certain phrases.
  • [0052]
    This training may use different statistical alignment models such as the IBM model 1, the HMM, and/or the IBM model 4. This constraint may also be integrated by constraining the set of considered alignments in the expectation maximization algorithm. This constraint may also improve the alignment quality of the surrounding words. For example, there may be a first order dependence among the alignment positions of the HMM and model for alignment models.
  • [0053]
    Some exemplary results are provided to explain the concepts. The results are based on a Chinese to English translation which was done in 2003. Table 1 provides statistics on the training, development and test corpus that was used. There are four reference translations, from the training corpus (train small, train large, dev and test.)
  • [0000]
    TABLE 1
    Characteristics of training corpus (Train),
    development corpus (Dev), test corpus (Test).
    Chinese English
    Train Segments 5 109
    (small) Words 89 121 111 251
    Train Segments  6.9 M
    (large) Words 170 M 157 M
    Dev Segments 935
    Words 27 012 27.6K-30.1K
    Test Segments 878
    Words 24 540 25.3K-28.6K
  • [0054]
    The system uses a subset of 128,000 sentences from the large parallel corpus to generate the translation component works-annotated corpus. Based on this corpus, 264,488 Chinese substrings can be translated using any of the rule based translation component, suggesting altogether approximately 364,000 translations. 60,589 of those translations, or 16.6%, also occur in the corresponding target language; called positive instances.
  • [0055]
    A review of these annotations shows that positive instances of the automatic corpus annotation are rarely incorrectly annotated, on the other hand, negative instances are much more frequent due to the existence of sentence alignment errors, and insufficient recall of the translation component.
  • [0056]
    For evaluation purposes, the test corpus was annotated in the same way as the training database. The test corpus is perfectly sentence aligned, and therefore there are no wrong negative instances due to alignment. In the test corpus, there are 2529 substrings that the translation component can translate, and when it does, it suggests 3651 translations of which 1287 (35.3%) also occur in any of the four references.
  • [0057]
    Using that annotated training corpus, the maximum entropy classifier described above is trained. Table 2 provides the results of this classifier for the development Corp. this for various training corpus sizes. This experiment uses Pmin=0.2.
  • [0000]
    TABLE 2
    Quality of classifier trained on the automatically annotated corpus
    Strict Loose
    # Segments Errors[%] Precision [%] Recall[%] Precision[%]
    1,000 18 79 65 88
    2,000 17 85 63 90
    4,000 16 86 67 91
    8,000 14 88 70 92
    16,000 13 89 71 94
    32,000 11 92 75 95
    64,000 9 94 78 97
    128,000 8 95 80 97
    (Errors[%]: error rate of classifier (percentage of suggested translations that are correct),
    (Strict) Precision[%]/Recall[%]: precision and recall of classifier,
    Loose Precision[%]: percentage of source language sub-strings where any of the suggested translations is correct).
  • [0058]
    In operation, a precision as high as 95% was eventually obtained with the recall of the person. See table 2 which shows the actual values. The column entitled “loose precision” provides a percentage of source language substrings where any of the suggested translations also occur in the references. Eventually the precision of 97% was achieved. This means that about 3% of the Chinese substrings for which a translation were not correct.
  • [0059]
    Word alignment that is computed by the statistical alignment models may be used to train the phrase based translation models, on those parts of the text where the automatic corpus annotation detects a translation. The automatic corpus annotation may be a very high precision, and can be used to improve the translation. One aspect, therefore, may improve general word alignment quality using the information in the translation component induced word alignment, in the statistical word alignment training.
  • [0060]
    Although only a few embodiments have been disclosed in detail above, other modifications are possible, and this disclosure is intended to cover all such modifications, and most particularly, any modification which might be predictable to a person having ordinary skill in the art. For example, the above has described integration of rule based translation components. It should be noted that other components, such as statistical components and the like may select alternative translations that can be used. The probability assigned by the model can be an additional feature for the classifier.
  • [0061]
    Also, only those claims which use the words “means for” are intended to be interpreted under 35 USC 112, sixth paragraph. Moreover, no limitations from the specification are intended to be read into any claims, unless those limitations are expressly included in the claims.
Patentzitate
Zitiertes PatentEingetragen Veröffentlichungsdatum Antragsteller Titel
US4502128 *1. Juni 198226. Febr. 1985Hitachi, Ltd.Translation between natural languages
US4661924 *19. Juli 198528. Apr. 1987Hitachi, Ltd.Multiple-parts-of-speech disambiguating method and apparatus for machine translation system
US4800522 *13. Mai 198624. Jan. 1989Sharp Kabushiki KaishaBilingual translation system capable of memorizing learned words
US4814987 *19. Mai 198621. März 1989Sharp Kabushiki KaishaTranslation system
US5088038 *24. Mai 199011. Febr. 1992Kabushiki Kaisha ToshibaMachine translation system and method of machine translation
US5091876 *18. Dez. 198925. Febr. 1992Kabushiki Kaisha ToshibaMachine translation system
US5181163 *31. Juli 198919. Jan. 1993Hitachi, Ltd.Method and apparatus for generating and/or updating cooccurrence relation dictionary
US5295068 *19. März 199115. März 1994Fujitsu LimitedApparatus for registering private-use words in machine-translation/electronic-mail system
US5302132 *1. Apr. 199212. Apr. 1994Corder Paul RInstructional system and method for improving communication skills
US5387104 *7. Febr. 19947. Febr. 1995Corder; Paul R.Instructional system for improving communication skills
US5408410 *13. Apr. 199318. Apr. 1995Hitachi, Ltd.Method of and an apparatus for automatically evaluating machine translation system through comparison of their translation results with human translated sentences
US5495413 *21. Sept. 199327. Febr. 1996Sharp Kabushiki KaishaTranslation machine having a function of deriving two or more syntaxes from one original sentence and giving precedence to a selected one of the syntaxes
US5497319 *26. Sept. 19945. März 1996Trans-Link International Corp.Machine translation and telecommunications system
US5510981 *28. Okt. 199323. Apr. 1996International Business Machines CorporationLanguage translation apparatus and method using context-based translation models
US5724593 *24. Jan. 19973. März 1998International Language Engineering Corp.Machine assisted translation tools
US5864788 *26. Mai 199526. Jan. 1999Sharp Kabushiki KaishaTranslation machine having a function of deriving two or more syntaxes from one original sentence and giving precedence to a selected one of the syntaxes
US5867811 *17. Juni 19942. Febr. 1999Canon Research Centre Europe Ltd.Method, an apparatus, a system, a storage device, and a computer readable medium using a bilingual database including aligned corpora
US5870706 *10. Apr. 19969. Febr. 1999Lucent Technologies, Inc.Method and apparatus for an improved language recognition system
US5893134 *21. Mai 19966. Apr. 1999Canon Europa N.V.Aligning source texts of different natural languages to produce or add to an aligned corpus
US6018617 *31. Juli 199725. Jan. 2000Advantage Learning Systems, Inc.Test generating and formatting system
US6031984 *9. März 199829. Febr. 2000I2 Technologies, Inc.Method and apparatus for optimizing constraint models
US6032111 *23. Juni 199729. Febr. 2000At&T Corp.Method and apparatus for compiling context-dependent rewrite rules and input strings
US6047252 *30. Juni 19974. Apr. 2000Kabushiki Kaisha ToshibaMachine translation method and source/target text display method
US6182014 *20. Nov. 199830. Jan. 2001Schlumberger Technology CorporationMethod and system for optimizing logistical operations in land seismic surveys
US6182027 *31. Juli 199830. Jan. 2001International Business Machines CorporationTranslation method and system
US6205436 *21. Aug. 199820. März 2001Citibank, N.A.Trusted agents for open electronic commerce where the transfer of electronic merchandise or electronic money is provisional until the transaction is finalized
US6206700 *14. Okt. 199427. März 2001Breakthrough To Literacy, Inc.Apparatus and method for interactive adaptive learning by an individual through at least one of a stimuli presentation device and a user perceivable display
US6223150 *29. Jan. 199924. Apr. 2001Sony CorporationMethod and apparatus for parsing in a spoken language translation system
US6356864 *23. Juli 199812. März 2002University Technology CorporationMethods for analysis and evaluation of the semantic content of a writing based on vector length
US6360196 *18. Mai 199919. März 2002Sharp Kabushiki KaishaMethod of and apparatus for retrieving information and storage medium
US6529865 *18. Okt. 19994. März 2003Sony CorporationSystem and method to compile instructions to manipulate linguistic structures into separate functions
US6535842 *9. Dez. 199918. März 2003Global Information Research And Technologies, LlcAutomatic bilingual translation memory system
US6691279 *30. März 199810. Febr. 2004Sanyo Electric Co., LtdDocument preparation method and machine translation device
US6857022 *29. Sept. 200015. Febr. 2005Worldlingo.Com Pty LtdTranslation ordering system
US6885985 *18. Dez. 200026. Apr. 2005Xerox CorporationTerminology translation for unaligned comparable corpora using category based translation probabilities
US6983239 *25. Okt. 20003. Jan. 2006International Business Machines CorporationMethod and apparatus for embedding grammars in a natural language understanding (NLU) statistical parser
US6996518 *3. Jan. 20017. Febr. 2006International Business Machines CorporationMethod and apparatus for automated measurement of quality for machine translation
US6996520 *14. Nov. 20037. Febr. 2006Transclick, Inc.Language translation system and method using specialized dictionaries
US6999925 *13. Nov. 200114. Febr. 2006International Business Machines CorporationMethod and apparatus for phonetic context adaptation for improved speech recognition
US7013262 *12. Febr. 200214. März 2006Sunflare Co., LtdSystem and method for accurate grammar analysis using a learners' model and part-of-speech tagged (POST) parser
US7016827 *3. Sept. 199921. März 2006International Business Machines CorporationMethod and system for ensuring robustness in natural language understanding
US7016977 *5. Nov. 199921. März 2006International Business Machines CorporationMethod and system for multilingual web server
US7024351 *21. Aug. 20014. Apr. 2006Microsoft CorporationMethod and apparatus for robust efficient parsing
US7171348 *7. Juli 200530. Jan. 2007Worldlingo.Com Pty LtdCommunication processing system
US7174289 *5. März 20016. Febr. 2007Oki Electric Industry Co., Ltd.Translating system and translating apparatus in which translatable documents are associated with permission to translate
US7177792 *31. Mai 200213. Febr. 2007University Of Southern CaliforniaInteger programming decoder for machine translation
US7191115 *17. Juni 200213. März 2007Microsoft CorporationStatistical method and apparatus for learning translation relationships among words
US7194403 *28. Sept. 200120. März 2007Fujitsu LimitedApparatus, method, and computer-readable medium for language translation
US7197451 *13. Juli 200027. März 2007Novell, Inc.Method and mechanism for the creation, maintenance, and comparison of semantic abstracts
US7319949 *27. Mai 200315. Jan. 2008Microsoft CorporationUnilingual translator
US7340388 *26. März 20034. März 2008University Of Southern CaliforniaStatistical translation using a large monolingual corpus
US7346487 *23. Juli 200318. März 2008Microsoft CorporationMethod and apparatus for identifying translations
US7346493 *25. März 200318. März 2008Microsoft CorporationLinguistically informed statistical models of constituent structure for ordering in sentence realization for a natural language generation system
US7349839 *27. Aug. 200225. März 2008Microsoft CorporationMethod and apparatus for aligning bilingual corpora
US7349845 *3. Sept. 200325. März 2008International Business Machines CorporationMethod and apparatus for dynamic modification of command weights in a natural language understanding system
US7496497 *18. Dez. 200324. Febr. 2009Taiwan Semiconductor Manufacturing Co., Ltd.Method and system for selecting web site home page by extracting site language cookie stored in an access device to identify directional information item
US7680646 *21. Dez. 200416. März 2010Xerox CorporationRetrieval method for translation memories containing highly structured documents
US7689405 *17. Mai 200230. März 2010Language Weaver, Inc.Statistical method for building a translation memory
US8145472 *12. Dez. 200627. März 2012John ShoreLanguage translation using a hybrid network of human and machine translators
US20020002451 *5. März 20013. Jan. 2002Tatsuya SukehiroTranslating system and translating apparatus
US20020013693 *23. Juni 199831. Jan. 2002Masaru FujiApparatus and method for controlling the display of a translation or dictionary searching process
US20020040392 *11. Apr. 20014. Apr. 2002Tonis KasvandExecution sets for generated logs
US20020046018 *11. Mai 200118. Apr. 2002Daniel MarcuDiscourse parsing and summarization
US20020046262 *16. Aug. 200118. Apr. 2002Joerg HeiligData access system and method with proxy and remote processing
US20030009322 *17. Mai 20029. Jan. 2003Daniel MarcuStatistical method for building a translation memory
US20030023423 *3. Juli 200230. Jan. 2003Kenji YamadaSyntax-based statistical translation model
US20040015342 *18. Febr. 200322. Jan. 2004Garst Peter F.Linguistic support for a recognizer of mathematical expressions
US20040024581 *28. März 20035. Febr. 2004Philipp KoehnStatistical machine translation
US20040030551 *27. März 200312. Febr. 2004Daniel MarcuPhrase to phrase joint probability model for statistical machine translation
US20040035055 *21. Aug. 200226. Febr. 2004Tianli ZhuSulfur control for fuel processing system for fuel cell power plant
US20040044530 *27. Aug. 20024. März 2004Moore Robert C.Method and apparatus for aligning bilingual corpora
US20040059708 *6. Dez. 200225. März 2004Google, Inc.Methods and apparatus for serving relevant advertisements
US20040068411 *22. Aug. 20038. Apr. 2004Philip ScanlanTranslation information segment
US20050021322 *25. Juli 200327. Jan. 2005Microsoft CorporationAdaptive machine translation
US20050021517 *26. Mai 200427. Jan. 2005Insightful CorporationExtended functionality for an inverse inference engine based web search
US20050026131 *31. Juli 20033. Febr. 2005Elzinga C. BretSystems and methods for providing a dynamic continual improvement educational environment
US20050033565 *2. Juli 200410. Febr. 2005Philipp KoehnEmpirical methods for splitting compound words with application to machine translation
US20050038643 *2. Juli 200417. Febr. 2005Philipp KoehnStatistical noun phrase translation
US20050055199 *19. Okt. 200110. März 2005Intel CorporationMethod and apparatus to provide a hierarchical index for a language model data structure
US20050060160 *16. Dez. 200317. März 2005Roh Yoon HyungHybrid automatic translation apparatus and method employing combination of rule-based method and translation pattern method, and computer-readable medium thereof
US20050075858 *6. Okt. 20037. Apr. 2005Microsoft CorporationSystem and method for translating from a source language to at least one target language utilizing a community of contributors
US20050086226 *6. Dez. 200421. Apr. 2005Albert KrachmanMethod and system for providing electronic discovery on computer databases and archives using statement analysis to detect false statements and recover relevant data
US20060004563 *30. Juni 20045. Jan. 2006Microsoft CorporationModule for creating a language neutral syntax representation using a language particular syntax tree
US20060015320 *15. Apr. 200519. Jan. 2006Och Franz JSelection and use of nonstatistical translation components in a statistical machine translation framework
US20060015323 *13. Juli 200419. Jan. 2006Udupa Raghavendra UMethod, apparatus, and computer program for statistical translation decoding
US20060018541 *29. Okt. 200426. Jan. 2006Microsoft CorporationAdaptation of exponential models
US20060020448 *29. Okt. 200426. Jan. 2006Microsoft CorporationMethod and apparatus for capitalizing text using maximum entropy
US20060041428 *20. Aug. 200423. Febr. 2006Juergen FritschAutomated extraction of semantic content and generation of a structured document from speech
US20070016400 *21. Juni 200518. Jan. 2007Radu SoricuttWeighted system of expressing language information using a compact notation
US20070016401 *12. Aug. 200518. Jan. 2007Farzad EhsaniSpeech-to-speech translation system with user-modifiable paraphrasing grammars
US20070033001 *3. Aug. 20058. Febr. 2007Ion MusleaIdentifying documents which form translated pairs, within a document collection
US20070050182 *25. Aug. 20051. März 2007Sneddon Michael VTranslation quality quantifying apparatus and method
US20080052061 *6. Aug. 200728. Febr. 2008Kim Young KilDomain-adaptive portable machine translation device for translating closed captions using dynamic translation resources and method thereof
US20080065478 *12. Sept. 200613. März 2008Microsoft CorporationElectronic coupon based service for enhancing content
US20090076792 *16. Juni 200819. März 2009Emil LtdText editing apparatus and method
US20090083023 *16. Juni 200626. März 2009George FosterMeans and Method for Adapted Language Translation
US20100017293 *17. Juli 200821. Jan. 2010Language Weaver, Inc.System, method, and computer program for providing multilingual text advertisments
US20100042398 *8. Okt. 200918. Febr. 2010Daniel MarcuBuilding A Translation Lexicon From Comparable, Non-Parallel Corpora
US20110029300 *28. Juli 20093. Febr. 2011Daniel MarcuTranslating Documents Based On Content
Nichtpatentzitate
Referenz
1 *Och, Franz Josef and Hermann Ney. 2002. Discriminative training and maximum entropy models for statistical machine translation. In Proc. of the 40th Annual Meeting of the Association for Computational Linguistics (ACL), Philadelphia, PA.
2 *Robert Frederking and Sergei Nirenburg. 1994. Three heads are better than one. In Proceedings of the 4th Conference on Applied Natural Language Processing, pages 95-100, Stuttgart, Germany.
3 *Yasuda, K., Sugaya, F., Takezawa, T., Yamamoto, S. and Yanagida, M. 2002 Automatic Machine Translation Selection Scheme to Output the Best Result, Proc. of LREC.
Referenziert von
Zitiert von PatentEingetragen Veröffentlichungsdatum Antragsteller Titel
US82141963. Juli 20023. Juli 2012University Of Southern CaliforniaSyntax-based statistical translation model
US82341068. Okt. 200931. Juli 2012University Of Southern CaliforniaBuilding a translation lexicon from comparable, non-parallel corpora
US8265923 *11. Mai 201011. Sept. 2012Xerox CorporationStatistical machine translation employing efficient parameter training
US829612722. März 200523. Okt. 2012University Of Southern CaliforniaDiscovery of parallel text portions in comparable collections of corpora and training using comparable texts
US83804861. Okt. 200919. Febr. 2013Language Weaver, Inc.Providing machine-generated translations and corresponding trust levels
US84335562. Nov. 200630. Apr. 2013University Of Southern CaliforniaSemi-supervised training for statistical word alignment
US846814926. Jan. 200718. Juni 2013Language Weaver, Inc.Multi-lingual online community
US85487942. Juli 20041. Okt. 2013University Of Southern CaliforniaStatistical noun phrase translation
US860072812. Okt. 20053. Dez. 2013University Of Southern CaliforniaTraining for a text-to-text application which uses string to tree conversion for training and decoding
US861538914. März 200824. Dez. 2013Language Weaver, Inc.Generation and exploitation of an approximate language model
US866672515. Apr. 20054. März 2014University Of Southern CaliforniaSelection and use of nonstatistical translation components in a statistical machine translation framework
US867656321. Juni 201018. März 2014Language Weaver, Inc.Providing human-generated and machine-generated trusted translations
US869430315. Juni 20118. Apr. 2014Language Weaver, Inc.Systems and methods for tuning parameters in statistical machine translation
US88254668. Juni 20072. Sept. 2014Language Weaver, Inc.Modification of annotated bilingual segment pairs in syntax-based machine translation
US88319284. Apr. 20079. Sept. 2014Language Weaver, Inc.Customizable machine translation service
US888651519. Okt. 201111. Nov. 2014Language Weaver, Inc.Systems and methods for enhancing machine translation post edit review processes
US888651729. Juni 201211. Nov. 2014Language Weaver, Inc.Trust scoring for language translation systems
US88865187. Aug. 200611. Nov. 2014Language Weaver, Inc.System and method for capitalizing machine translated text
US89429739. März 201227. Jan. 2015Language Weaver, Inc.Content page URL translation
US89430805. Dez. 200627. Jan. 2015University Of Southern CaliforniaSystems and methods for identifying parallel documents and sentence fragments in multilingual document collections
US899006428. Juli 200924. März 2015Language Weaver, Inc.Translating documents based on content
US912267415. Dez. 20061. Sept. 2015Language Weaver, Inc.Use of annotations in statistical machine translation
US915262226. Nov. 20126. Okt. 2015Language Weaver, Inc.Personalized machine translation via online adaptation
US921369410. Okt. 201315. Dez. 2015Language Weaver, Inc.Efficient online domain adaptation
US9501470 *7. Febr. 201322. Nov. 2016At&T Intellectual Property I, L.P.System and method for enriching spoken language translation with dialog acts
US953016128. Febr. 201427. Dez. 2016Ebay Inc.Automatic extraction of multilingual dictionary items from non-parallel, multilingual, semi-structured data
US956952628. Febr. 201414. Febr. 2017Ebay Inc.Automatic machine translation using user feedback
US979872026. Okt. 200924. Okt. 2017Ebay Inc.Hybrid machine translation
US980503113. Dez. 201631. Okt. 2017Ebay Inc.Automatic extraction of multilingual dictionary items from non-parallel, multilingual, semi-structured data
US20100088085 *9. Apr. 20098. Apr. 2010Jae-Hun JeonStatistical machine translation apparatus and method
US20100179803 *26. Okt. 200915. Juli 2010AppTekHybrid machine translation
US20110282643 *11. Mai 201017. Nov. 2011Xerox CorporationStatistical machine translation employing efficient parameter training
US20120215520 *23. Febr. 201123. Aug. 2012Davis Janel RTranslation System
Klassifizierungen
US-Klassifikation704/2
Internationale KlassifikationG06F17/28
UnternehmensklassifikationG06F17/2818
Europäische KlassifikationG06F17/28D2
Juristische Ereignisse
DatumCodeEreignisBeschreibung
3. Juni 2008ASAssignment
Owner name: UNIVERSITY OF SOUTHERN CALIFORNIA, CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:UNIVERSITY OF SOUTHERN CALIFORNIA;OCH, FRANZ JOSEF;REEL/FRAME:021044/0638
Effective date: 20050809
17. Juli 2008ASAssignment
Owner name: UNIVERSITY OF SOUTHERN CALIFORNIA, CALIFORNIA
Free format text: THIS CORR;ASSIGNOR:OCH, FRANZ JOSEF;REEL/FRAME:021277/0264
Effective date: 20050809
Owner name: UNIVERSITY OF SOUTHERN CALIFORNIA, CALIFORNIA
Free format text: THIS CORRECTIVE ASSIGNMENT CORRECTS THE NAME OF THE ASSIGNOR WHICH IS FRANZ JOSEF OCH. THE PREVIOUSLY FILED ASSIGNMENT COVER SHEET INCORRECTLY LISTS THE ASSIGNORS AS "FRANZ JOSEF OCH" AS REFLECTED BY BOTH THE ASSIGNMENT AND THE NEW ASSIGNMENT COVER SHEET;ASSIGNOR:OCH, FRANZ JOSEF;REEL/FRAME:021277/0264
Effective date: 20050809